Tag: out-of-distribution generalization

A sample-aware test-time adaptation framework selectively updates internal generative feature representations to resolve medical image translation errors caused by clinical ...

Researchers show that lexical and embedding-based drift measures between synthetic and real clinical text moderately predict how much de-identification models ...

A weighted training method that distinguishes likely shortcut examples improved AI robustness across four image and language benchmarks without adding ...

A physics-informed residual network reconstructed aneurysm blood-flow fields more accurately than standard machine-learning models, with its strongest gains appearing in ...

Researchers at GFZ Potsdam have developed a three-step method that extrapolates a trained neural network's own weight sensitivities to build ...

Researchers have developed CCOT, a condition-controlled optimal transport framework that predicts single-cell responses to drug perturbations, including to compounds never ...